Sketch: Generative Architecture Platform
Apps & Workflows · SCI-Arc · 2023
How does AI change the architect's agency during early ideation? Sketch answers by giving the site a vote. Address in, massing and facade out, both shaped by slope, sun and trees.
A year after Midjourney, architects were generating concept imagery faster than any studio could draw it. All of it flat. The images looked resolved and none of them could be built, because there was no model underneath. Sketch was an attempt to move that speed into three dimensions, where a site and a sun angle still apply.
Thesis
Sketch was developed at SCI-Arc as a thesis project. The field was split between tools that optimized numbers and tools that generated images.
Watch: Sketch: Generative Architecture Platform
The question driving it was whether site analysis and image generation could share one workflow, so the concept phase could move at the speed of an idea without leaving feasibility behind.
Address to massing
You enter an address and draw the site boundaries on a map. The platform reads slope steepness, flags features worth keeping like native trees, and runs a year-round solar analysis. Objectives are yours to set. In Los Angeles you might ask it to minimize sun exposure, and the massing options it returns change accordingly.
Watch: Sketch: Generative Architecture Platform
Watch: Sketch: Generative Architecture Platform
Prompt to facade
The prompt sets the visual language. The solar data sets where it lands. Windows go into the sheltered zones with softer light, and solid walls or panels take the surfaces in direct sun. The same prompt produces a different facade on a different site.
Watch: Sketch: Generative Architecture Platform
Sketch is the Co-Lab thesis built into software, three months of work from the site solver to the Unreal export. A site can propose its own massing. A prompt can produce a facade that already knows where the sun hits.
Generation and constraint do not have to be separate steps.